openclaw-comfyui-imagegenerate

通过本地 ComfyUI 生成图片并发送到指定飞书会话，成功发送后无回复文字，保持界面简洁。

Security Scan

The skill's code, instructions, and required resources are consistent with its stated purpose (generate images via a local ComfyUI instance and silently send them to a Feishu conversation).

Purpose & Capability

Name/description claim: generate via local ComfyUI and send to Feishu silently. The runtime code (on_call_skill + draw.py) only invokes a local draw.py, interacts with ComfyUI on 127.0.0.1:8188, and uses OpenClaw's message.send to deliver the image to Feishu — these match the declared purpose. No unrelated credentials or remote services are requested.

Instruction Scope

SKILL.md instructs running draw.py, capturing a single stdout path, and sending that file to the Feishu target from context. draw.py only reads zimage-api.json, talks to localhost ComfyUI HTTP/WebSocket endpoints, saves an image under its output_images folder, and prints the image path — all within the scope of image generation and delivery. Note: success is intentionally silent (returns "NO_REPLY"); error paths return human-readable errors.

Install Mechanism

No install spec (instruction-only) and no external download steps. draw.py relies on standard urllib and a websocket client library; the absence of an install step means those dependencies must already exist in the environment, but there is no high-risk installer or remote code download.

Credentials

Skill declares no required env vars or credentials. It expects platform-provided Feishu context (feishu_chat_id / feishu_user_id) and that the OpenClaw environment has the Feishu bot permissions to send messages. The only network access is to localhost (ComfyUI). There are no unrelated secrets requested.

Persistence & Privilege

always:false (no forced persistence). The skill writes generated images to its own output_images directory (local, expected). It does not modify other skills or global agent configuration.

Assessment

This skill appears coherent for its purpose, but consider these operational points before installing: (1) It expects a local ComfyUI service at 127.0.0.1:8188 and the websocket client library to be present; if those are not available the skill will fail. (2) Generated images are written to the skill directory (output_images) — ensure you are comfortable with local storage of generated content. (3) The prompt text is sent to Feishu along with the image; avoid sending sensitive prompts you don't want delivered to that chat. (4) The skill hardcodes a working directory in the example SKILL.md — verify the actual runtime path used by your OpenClaw deployment so draw.py and zimage-api.json are reachable. If you need more assurance, request the developer to include an explicit install spec (or dependency list) and to remove or parameterize the hardcoded path.

name: comfyui-feishu-draw description: 调用本地 ComfyUI 生成图片并直接发送到飞书当前会话。任务完成后直接结束，严禁返回任何总结或确认文字。 核心任务：生成图片并发送飞书 🚀 技能实现 Python

import subprocess import os from openclaw.tools import message

def on_call_skill(prompt, context): """ 核心逻辑：生成图片 -> 获取路径 -> 发送飞书 -> 静默退出 """ # 1. 运行本地绘制脚本获取图片路径 skill_dir = "/home/coder/.openclaw/workspace/skills/comfyui-img-gen"

try:
    # 运行 draw.py，stdout 仅返回生成的绝对路径
    result = subprocess.run(
        ["python3", "draw.py", prompt],
        cwd=skill_dir,
        capture_output=True,
        text=True,
        timeout=60
    )
    
    image_path = result.stdout.strip()
    
    if result.returncode != 0 or not os.path.exists(image_path):
        return f"❌ 图片生成失败: {result.stderr}"

    # 2. 识别飞书发送目标（群聊或私聊）
    target = None
    if context.get('feishu_chat_id'):
        target = f"chat:{context['feishu_chat_id']}"
    elif context.get('feishu_user_id'):
        target = f"user:{context['feishu_user_id']}"
    
    if not target:
        return "❌ 未检测到有效的飞书会话上下文"

    # 3. 调用 OpenClaw 内置工具发送消息
    message.send(
        channel="feishu",
        target=target,
        message=f"🎨 已为您生成图片：{prompt}",
        path=image_path
    )
    
    # 4. 关键：成功发送后返回 NO_REPLY 避免额外文字
    # OpenClaw 会将 NO_REPLY 视为静默指令，不发送任何回复
    return "NO_REPLY"

except Exception as e:
    return f"❌ 任务执行异常: {str(e)}"

⚙️ 核心依赖 本地服务: ComfyUI (127.0.0.1:8188) 处于启动状态。

本地文件: draw.py 和 zimage-api.json 需在 skill_dir 指定目录中。

飞书权限: 机器人需具备 im:message:send_as_bot 权限。

📝 静默返回说明 当技能成功发送图片到飞书后，返回 "NO_REPLY" 字符串。 OpenClaw 会将其识别为静默指令，不会在聊天界面显示任何回复文字。 这样用户只会在飞书中收到图片，而不会在技能调用界面看到额外的成功提示。